tornavis/source/blender/blenkernel/BKE_node_runtime.hh

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

568 lines
17 KiB
C++
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later */
#pragma once
#include <memory>
#include <mutex>
#include "BLI_multi_value_map.hh"
#include "BLI_utility_mixins.hh"
#include "BLI_vector.hh"
#include "DNA_node_types.h"
#include "BKE_node.h"
struct bNode;
struct bNodeSocket;
struct bNodeTree;
struct bNodeType;
namespace blender::nodes {
struct FieldInferencingInterface;
2022-05-30 16:27:59 +02:00
class NodeDeclaration;
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
struct GeometryNodesLazyFunctionGraphInfo;
} // namespace blender::nodes
namespace blender::bke {
class bNodeTreeRuntime : NonCopyable, NonMovable {
public:
/**
* Keeps track of what changed in the node tree until the next update.
* Should not be changed directly, instead use the functions in `BKE_node_tree_update.h`.
* #eNodeTreeChangedFlag.
*/
uint32_t changed_flag = 0;
/**
* A hash of the topology of the node tree leading up to the outputs. This is used to determine
* of the node tree changed in a way that requires updating geometry nodes or shaders.
*/
uint32_t output_topology_hash = 0;
/**
* Used to cache run-time information of the node tree.
* #eNodeTreeRuntimeFlag.
*/
uint8_t runtime_flag = 0;
/** Information about how inputs and outputs of the node group interact with fields. */
std::unique_ptr<nodes::FieldInferencingInterface> field_inferencing_interface;
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
/**
* For geometry nodes, a lazy function graph with some additional info is cached. This is used to
* evaluate the node group. Caching it here allows us to reuse the preprocessed node tree in case
* its used multiple times.
*/
std::mutex geometry_nodes_lazy_function_graph_info_mutex;
std::unique_ptr<nodes::GeometryNodesLazyFunctionGraphInfo>
geometry_nodes_lazy_function_graph_info;
/**
* Protects access to all topology cache variables below. This is necessary so that the cache can
* be updated on a const #bNodeTree.
*/
std::mutex topology_cache_mutex;
bool topology_cache_is_dirty = true;
bool topology_cache_exists = false;
/**
* Under some circumstances, it can be useful to use the cached data while editing the
* #bNodeTree. By default, this is protected against using an assert.
*/
mutable std::atomic<int> allow_use_dirty_topology_cache = 0;
/** Only valid when #topology_cache_is_dirty is false. */
Vector<bNode *> nodes;
Vector<bNodeLink *> links;
Vector<bNodeSocket *> sockets;
Vector<bNodeSocket *> input_sockets;
Vector<bNodeSocket *> output_sockets;
MultiValueMap<const bNodeType *, bNode *> nodes_by_type;
Vector<bNode *> toposort_left_to_right;
Vector<bNode *> toposort_right_to_left;
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
Vector<bNode *> group_nodes;
bool has_link_cycle = false;
bool has_undefined_nodes_or_sockets = false;
bNode *group_output_node = nullptr;
};
/**
* Run-time data for every socket. This should only contain data that is somewhat persistent (i.e.
* data that lives longer than a single depsgraph evaluation + redraw). Data that's only used in
* smaller scopes should generally be stored in separate arrays and/or maps.
*/
class bNodeSocketRuntime : NonCopyable, NonMovable {
public:
/**
* References a socket declaration that is owned by `node->declaration`. This is only runtime
* data. It has to be updated when the node declaration changes. Access can be allowed by using
* #AllowUsingOutdatedInfo.
*/
const SocketDeclarationHandle *declaration = nullptr;
/** #eNodeTreeChangedFlag. */
uint32_t changed_flag = 0;
/** Only valid when #topology_cache_is_dirty is false. */
Vector<bNodeLink *> directly_linked_links;
Vector<bNodeSocket *> directly_linked_sockets;
Vector<bNodeSocket *> logically_linked_sockets;
Vector<bNodeSocket *> logically_linked_skipped_sockets;
bNode *owner_node = nullptr;
bNodeSocket *internal_link_input = nullptr;
int index_in_node = -1;
int index_in_all_sockets = -1;
int index_in_inout_sockets = -1;
};
/**
* Run-time data for every node. This should only contain data that is somewhat persistent (i.e.
* data that lives longer than a single depsgraph evaluation + redraw). Data that's only used in
* smaller scopes should generally be stored in separate arrays and/or maps.
*/
class bNodeRuntime : NonCopyable, NonMovable {
public:
/**
* Describes the desired interface of the node. This is run-time data only.
* The actual interface of the node may deviate from the declaration temporarily.
* It's possible to sync the actual state of the node to the desired state. Currently, this is
* only done when a node is created or loaded.
*
* In the future, we may want to keep more data only in the declaration, so that it does not have
* to be synced to other places that are stored in files. That especially applies to data that
* can't be edited by users directly (e.g. min/max values of sockets, tooltips, ...).
*
* The declaration of a node can be recreated at any time when it is used. Caching it here is
* just a bit more efficient when it is used a lot. To make sure that the cache is up-to-date,
* call #nodeDeclarationEnsure before using it.
*
* Currently, the declaration is the same for every node of the same type. Going forward, that is
* intended to change though. Especially when nodes become more dynamic with respect to how many
* sockets they have.
*/
NodeDeclarationHandle *declaration = nullptr;
/** #eNodeTreeChangedFlag. */
uint32_t changed_flag = 0;
/** Only valid if #topology_cache_is_dirty is false. */
Vector<bNodeSocket *> inputs;
Vector<bNodeSocket *> outputs;
Vector<bNodeLink *> internal_links;
Map<StringRefNull, bNodeSocket *> inputs_by_identifier;
Map<StringRefNull, bNodeSocket *> outputs_by_identifier;
int index_in_tree = -1;
bool has_linked_inputs = false;
bool has_linked_outputs = false;
bNodeTree *owner_tree = nullptr;
};
namespace node_tree_runtime {
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
/**
* Is executed when the depsgraph determines that something in the node group changed that will
* affect the output.
*/
void handle_node_tree_output_changed(bNodeTree &tree_cow);
class AllowUsingOutdatedInfo : NonCopyable, NonMovable {
private:
const bNodeTree &tree_;
public:
AllowUsingOutdatedInfo(const bNodeTree &tree) : tree_(tree)
{
tree_.runtime->allow_use_dirty_topology_cache.fetch_add(1);
}
~AllowUsingOutdatedInfo()
{
tree_.runtime->allow_use_dirty_topology_cache.fetch_sub(1);
}
};
inline bool topology_cache_is_available(const bNodeTree &tree)
{
if (!tree.runtime->topology_cache_exists) {
return false;
}
if (tree.runtime->allow_use_dirty_topology_cache.load() > 0) {
return true;
}
if (tree.runtime->topology_cache_is_dirty) {
return false;
}
return true;
}
inline bool topology_cache_is_available(const bNode &node)
{
const bNodeTree *ntree = node.runtime->owner_tree;
if (ntree == nullptr) {
return false;
}
return topology_cache_is_available(*ntree);
}
inline bool topology_cache_is_available(const bNodeSocket &socket)
{
const bNode *node = socket.runtime->owner_node;
if (node == nullptr) {
return false;
}
return topology_cache_is_available(*node);
}
} // namespace node_tree_runtime
} // namespace blender::bke
/* -------------------------------------------------------------------- */
/** \name #bNodeTree Inline Methods
* \{ */
inline blender::Span<bNode *> bNodeTree::nodes_by_type(const blender::StringRefNull type_idname)
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->nodes_by_type.lookup(nodeTypeFind(type_idname.c_str()));
}
inline blender::Span<const bNode *> bNodeTree::nodes_by_type(
const blender::StringRefNull type_idname) const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->nodes_by_type.lookup(nodeTypeFind(type_idname.c_str()));
}
inline blender::Span<const bNode *> bNodeTree::toposort_left_to_right() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->toposort_left_to_right;
}
inline blender::Span<const bNode *> bNodeTree::toposort_right_to_left() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->toposort_right_to_left;
}
inline blender::Span<const bNode *> bNodeTree::all_nodes() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->nodes;
}
inline blender::Span<bNode *> bNodeTree::all_nodes()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->nodes;
}
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
inline blender::Span<const bNode *> bNodeTree::group_nodes() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->group_nodes;
}
inline blender::Span<bNode *> bNodeTree::group_nodes()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->group_nodes;
}
inline bool bNodeTree::has_link_cycle() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->has_link_cycle;
}
inline bool bNodeTree::has_undefined_nodes_or_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->has_undefined_nodes_or_sockets;
}
inline const bNode *bNodeTree::group_output_node() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->group_output_node;
}
inline blender::Span<const bNodeSocket *> bNodeTree::all_input_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->input_sockets;
}
inline blender::Span<bNodeSocket *> bNodeTree::all_input_sockets()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->input_sockets;
}
inline blender::Span<const bNodeSocket *> bNodeTree::all_output_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->output_sockets;
}
inline blender::Span<bNodeSocket *> bNodeTree::all_output_sockets()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->output_sockets;
}
inline blender::Span<const bNodeSocket *> bNodeTree::all_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->sockets;
}
inline blender::Span<bNodeSocket *> bNodeTree::all_sockets()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->sockets;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name #bNode Inline Methods
* \{ */
inline blender::Span<bNodeSocket *> bNode::input_sockets()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->inputs;
}
inline blender::Span<bNodeSocket *> bNode::output_sockets()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->outputs;
}
inline blender::Span<const bNodeSocket *> bNode::input_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->inputs;
}
inline blender::Span<const bNodeSocket *> bNode::output_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->outputs;
}
inline bNodeSocket &bNode::input_socket(int index)
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->inputs[index];
}
inline bNodeSocket &bNode::output_socket(int index)
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->outputs[index];
}
inline const bNodeSocket &bNode::input_socket(int index) const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->inputs[index];
}
inline const bNodeSocket &bNode::output_socket(int index) const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->outputs[index];
}
inline const bNodeSocket &bNode::input_by_identifier(blender::StringRef identifier) const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->inputs_by_identifier.lookup_as(identifier);
}
inline const bNodeSocket &bNode::output_by_identifier(blender::StringRef identifier) const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->outputs_by_identifier.lookup_as(identifier);
}
2022-09-07 10:24:34 +02:00
inline const bNodeTree &bNode::owner_tree() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->owner_tree;
}
inline blender::StringRefNull bNode::label_or_name() const
{
if (this->label[0] == '\0') {
return this->name;
}
return this->label;
}
inline bool bNode::is_muted() const
{
return this->flag & NODE_MUTED;
}
inline bool bNode::is_reroute() const
{
return this->type == NODE_REROUTE;
}
inline bool bNode::is_frame() const
{
return this->type == NODE_FRAME;
}
inline bool bNode::is_group() const
{
return ELEM(this->type, NODE_GROUP, NODE_CUSTOM_GROUP);
}
inline bool bNode::is_group_input() const
{
return this->type == NODE_GROUP_INPUT;
}
inline bool bNode::is_group_output() const
{
return this->type == NODE_GROUP_OUTPUT;
}
inline blender::Span<const bNodeLink *> bNode::internal_links_span() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->internal_links;
}
inline const blender::nodes::NodeDeclaration *bNode::declaration() const
{
return this->runtime->declaration;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name #bNodeLink Inline Methods
* \{ */
inline bool bNodeLink::is_muted() const
{
return this->flag & NODE_LINK_MUTED;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name #bNodeSocket Inline Methods
* \{ */
inline int bNodeSocket::index() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->index_in_node;
}
inline int bNodeSocket::index_in_tree() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->index_in_all_sockets;
}
inline bool bNodeSocket::is_available() const
{
return (this->flag & SOCK_UNAVAIL) == 0;
}
inline bNode &bNodeSocket::owner_node()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->owner_node;
}
inline const bNodeTree &bNodeSocket::owner_tree() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->owner_node->runtime->owner_tree;
}
inline blender::Span<const bNodeSocket *> bNodeSocket::logically_linked_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->logically_linked_sockets;
}
inline blender::Span<const bNodeLink *> bNodeSocket::directly_linked_links() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->directly_linked_links;
}
inline blender::Span<bNodeLink *> bNodeSocket::directly_linked_links()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->directly_linked_links;
}
inline blender::Span<const bNodeSocket *> bNodeSocket::directly_linked_sockets() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->directly_linked_sockets;
}
inline blender::Span<bNodeSocket *> bNodeSocket::directly_linked_sockets()
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return this->runtime->directly_linked_sockets;
}
inline bool bNodeSocket::is_directly_linked() const
{
return !this->directly_linked_links().is_empty();
}
inline bool bNodeSocket::is_logically_linked() const
{
return !this->logically_linked_sockets().is_empty();
}
inline const bNodeSocket *bNodeSocket::internal_link_input() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
BLI_assert(this->in_out == SOCK_OUT);
return this->runtime->internal_link_input;
}
template<typename T> const T *bNodeSocket::default_value_typed() const
{
return static_cast<const T *>(this->default_value);
}
inline bool bNodeSocket::is_input() const
{
return this->in_out == SOCK_IN;
}
inline bool bNodeSocket::is_output() const
{
return this->in_out == SOCK_OUT;
}
inline bool bNodeSocket::is_multi_input() const
{
return this->flag & SOCK_MULTI_INPUT;
}
inline const bNode &bNodeSocket::owner_node() const
{
BLI_assert(blender::bke::node_tree_runtime::topology_cache_is_available(*this));
return *this->runtime->owner_node;
}
/** \} */